Speaker Identification, Gmm Performance Method

It's one of the important research fields of speech recognition that using information extracted from the speech signal to perform speaker recognition. Based on the context of speech signal, speaker recognition can be divided into text-dependent and text-independent. It is very attracting of text-independent speaker identification due to more flexible and widely application.In the text-dependent speaker recognition, the GMM shifts the problem of speaker recognition to the problem of the estimation of distribution of training data. Thus, it divides more complex problems of data training and pattern matching into some simple problems, such as parameter estimation and computation of probability. Also, GMM has characteristics of simple, flexible and robust. So it is the-state-of-art in text-independent speaker recognition.However, GMM relies on training data to make sure of reliably estimation of parameters. Firstly, full covariance matrices can't be estimated reliably due to dramatic increasing of number of parameters while the dimension of feature vector is high. Secondly, though diagonal covariance matrices can be estimated reliably due to much less parameters than full covariance matrices, there is an implied assumption that inter-elements of feature are uncorrelated. In most application, this assumption is not reasonable. Finally, in the standard MAP algorithm, only mean parameters can be adapt from the UBM to the target GMM due to limited training data although covariance parameters carry speaker information too.In order to improve the performance of the text-independent speaker recognition system, this thesis study on three parts mentioned above to improve the GMM with limited training data.Firstly, this thesis presents a new framework in which full covariance matrices can be estimated reliably based on sharing correlation matrices. This framework can overcome the problem of unreliably estimating of full covariance matrices due to many parameters. In this framework, the GMM can describe the correlation of inter-elements of feature vectors by using the full covariance matrix, and the number of parameters is not too many to be estimated.Secondly, although we can decorrelate the feature vector in the feature space, but...